text-embedding-3-large not appearing in AI Foundry Chat Playground "Add your data" dropdown despite successful deployment in US East 2

Question

I am experiencing a UI limitation in the Azure AI Foundry Chat Playground where I cannot select my deployed text-embedding-3-large model when connecting an Azure AI Search data source.

Environment Details:

Region: US East 2

Search Service Tier: (e.g., Basic/S1)

Model Deployments: * gpt-35-turbo (Succeeded)

text-embedding-3-large-legal-ai (Version 1, Global Standard, Succeeded)

  `text-embedding-ada-002` (Version 2, Global Standard, Succeeded)
  

The Problem:

I created an Azure AI Search index (rag-prefix) with 3072 dimensions using text-embedding-3-large for legal document chunking.

In the Chat Playground, I select "Add your data" and choose my AI Search resource and the rag-prefix index.

When prompted to select an embedding model to enable vector search, the dropdown only displays text-embedding-ada-002.

My text-embedding-3-large deployment is completely missing from this specific dropdown, even though it is active and working in the indexing pipeline.

Observations:

If I deploy text-embedding-ada-002 (which is Version 2), it appears immediately.

It seems the Playground UI filters out text-embedding-3-large (which is labeled as Version 1), possibly due to a version-sorting or capability-mapping bug in the "Add your data" wizard.

• Because I cannot select the correct model, I am forced to use 1536-dimension models, which are incompatible with my high-precision 3072-dimension legal index.I am experiencing a UI limitation in the Azure AI Foundry Chat Playground where I cannot select my deployed text-embedding-3-large model when connecting an Azure AI Search data source. Environment Details:
  • Region: US East 2
  • Search Service Tier: (e.g., Basic/S1)
  • Model Deployments: * gpt-35-turbo (Succeeded)
    • text-embedding-3-large-legal-ai (Version 1, Global Standard, Succeeded)
    • text-embedding-ada-002 (Version 2, Global Standard, Succeeded)
  The Problem:
  1. I created an Azure AI Search index (rag-prefix) with 3072 dimensions using text-embedding-3-large for legal document chunking.
  2. In the Chat Playground, I select "Add your data" and choose my AI Search resource and the rag-prefix index.
  3. When prompted to select an embedding model to enable vector search, the dropdown only displays text-embedding-ada-002.
  4. My text-embedding-3-large deployment is completely missing from this specific dropdown, even though it is active and working in the indexing pipeline.
  Observations:
  • If I deploy text-embedding-ada-002 (which is Version 2), it appears immediately.
  • It seems the Playground UI filters out text-embedding-3-large (which is labeled as Version 1), possibly due to a version-sorting or capability-mapping bug in the "Add your data" wizard.
  • Because I cannot select the correct model, I am forced to use 1536-dimension models, which are incompatible with my high-precision 3072-dimension legal index.

Answer 1

Hi lahcene dergham,

This is a known and common confusion in Azure AI Foundry, and in most cases it is not a deployment or quota issue.

What you are seeing is mainly due to how the AI Foundry playground and the Add your data experience currently filter embedding models.

First, confirm that the model deployment itself is healthy. If text embedding 3 large shows as Succeeded in the Deployments section and works when you test it directly using the embeddings API, then the model is deployed correctly and usable. The model catalog page confirms that text embedding 3 large is a supported and active model in Azure AI Foundry. https://ai.azure.com/catalog/models/text-embedding-3-large

Second, understand why it does not appear in the Chat Playground Add your data dropdown. The Chat Playground currently shows only a limited subset of embedding models when you connect Azure AI Search as a data source. At the moment, the UI primarily exposes older or preview aligned models such as text embedding ada 002. Even though text embedding 3 large works for indexing and API calls, it may not appear in this specific dropdown due to UI capability mapping and version filtering. This behavior has been reported by multiple users and matches what you are experiencing.

Third, check how your Azure AI Search index was created. If your search index was created with 3072 dimensions using text embedding 3 large, that setup is valid. However, the Chat Playground Add your data flow can still block model selection even when the index itself is correct. This does not impact programmatic usage or ingestion pipelines, only the playground UI.

Fourth, use the model outside the playground UI. Text embedding 3 large works correctly when used through the embeddings API or for integrated vectorization during indexing, as long as the deployment name, endpoint, and API version are correct. Many users successfully use it for RAG pipelines even though it does not show up in the Chat Playground dropdown. https://learn.microsoft.com/azure/search/vector-search-integrated-vectorization-ai-studio

Finally, test and validate the setup in code. If the model generates embeddings correctly via REST or SDK calls and your Azure AI Search queries work as expected, then your environment is correctly configured. The limitation you are seeing is specific to the current portal experience and not a blocker for real usage.

In short, text embedding 3 large is supported and works, but it may not appear in the AI Foundry Chat Playground Add your data model list due to current UI limitations. Using the model directly through APIs or indexing workflows is the recommended path for now.

Hope this clears things up. Let me know if you have more questions.

Thankyou!

Answer 2

In the Chat Playground “Add your data” flow, only a specific, whitelisted set of embedding models is currently supported. The UI isn’t showing all successful deployments from the Azure OpenAI / Foundry resource; it filters to the models that the playground integration explicitly supports.

From the documented behavior of Foundry and Azure AI Search integrations:

• When adding knowledge to an agent workflow or using integrated vectorization, the supported embedding models are:
  • text-embedding-3-small
  • text-embedding-3-large
  • text-embedding-ada-002
  • Cohere-embed-v3-english
  • Cohere-embed-v3-multilingual
• For Azure AI Search integrated vectorization, the supported Azure OpenAI models are:
  • text-embedding-ada-002
  • text-embedding-3-small
  • text-embedding-3-large

However, the quickstart and portal wizard for vector search and “Add your data” explicitly list only certain combinations depending on the resource type and deployment style (for example, some Cohere models are only supported as serverless deployments). This means the playground relies on internal capability checks rather than just “any successful deployment.”

Given that:

1. text-embedding-3-large is a supported model in principle, and is available in eastus2 according to the model availability tables.
2. The model is already working in the indexing pipeline (so the deployment itself is valid).
3. The dropdown shows text-embedding-ada-002 but not text-embedding-3-large for the same resource and region.

The behavior aligns with a limitation/bug in the current Chat Playground “Add your data” wizard rather than a configuration issue in the resource or region:

• The playground UI is not correctly enumerating all supported embedding deployments for the Azure AI Search data source scenario, and is only exposing text-embedding-ada-002 in that picker.
• Because the wizard enforces a single embedding selection and validates it against its own supported list, it effectively blocks using a 3072‑dimension text-embedding-3-large index from that UI.

Practical implications and workarounds based on the documented capabilities:

1. text-embedding-3-large is supported for Azure AI Search integrated vectorization and is available in eastus2, so it can be used programmatically (for example, via REST/SDK or custom app) even if the Chat Playground wizard does not expose it.
2. To keep using the 3072‑dimension index with text-embedding-3-large, use a custom client (web app, API, or Promptflow/agent workflow) that:
   • Calls Azure AI Search directly with the 3072‑dimension vectors.
   • Uses the text-embedding-3-large deployment for query embeddings outside of the Chat Playground UI.
3. If the Chat Playground experience is required, the only embedding model currently exposed by that specific dropdown in this scenario is text-embedding-ada-002, which is limited to 1536 dimensions, so it will not match the existing 3072‑dimension index.

Because the model and region are supported but the UI does not surface the deployment, this is a product issue rather than a misconfiguration. The next step is to open a support ticket with Azure support, referencing that:

• text-embedding-3-large is deployed and working for indexing in eastus2.
• The Chat Playground “Add your data” → Azure AI Search data source → embedding model dropdown only shows text-embedding-ada-002 and omits text-embedding-3-large.
• The index uses 3072 dimensions and cannot be queried correctly with 1536‑dimension embeddings.

Support can confirm the current product limitation and track it as a bug or feature gap in the Chat Playground wizard.

---

References:

• Generate embeddings for search queries and documents
• Quickstart: Vector search in the Azure portal
• Set up integrated vectorization in Azure AI Search using REST
• Azure OpenAI in Azure AI Foundry Models
• Foundry Models sold directly by Azure (azure-openai)